[Feat]: Inference Settings by Chat, not by model

[sotwi]

Description
Currently the inference settings are linked to the model instead of the chat.

Use Case
Let's say I want to use the same model for Question Answering and Story Generation. I'd probably want different system prompts as well as different settings (less randomness for QA more for Story Gen).

It would be useful to modify the inference settings for each individual chat regardless of the model.